Application of soft-computing techniques for statistical modeling and optimization of seed oil extraction using polar and non-polar solvents.

This research presents optimal factor evaluation for maximum seed oil (DESO) extraction by applying central composite design (CCD) based on Box-Behnken (BB) experimental design of response surface methodology (RSM) and Artificial neural network (ANN) on feed forward-back propagation (FFBP) of Levenberg Marquardt (LM) training algorithm. Polar solvents (ethanol and combination of methanol and chloroform (M/C)) and non-polar solvents (n-hexane) were used for the extraction. The RSM optimal predicted oil yields were 45.21%, 38.61% and 30.87% while experimental values were 46.01%, 40.71% and 32.45% for n-hexane, ethanol and M/C respectively. The RSM optimum conditions were particle size of 450.67, 451.19 and 450.22μm, extraction time of 55.57, 55.16 and 56.11min and solute/solvent ratio of 0.19, 0.16 and 0.18 g/ml for n-hexane, ethanol and M/C respectively. The ANN-GA optimized conditions showed 5.14, 5.81 and 2.12 % higher DESO yields at 1.10, 0.26 and 0.65% smaller particle sizes, 5.47, 0.30 and 0.62 % faster extraction rate, and 24, 11.11 and 10% more solute requirement, for n-hexane, ethanol and M/C solvents respectively. The particle size was found to be the most significant factor. ANN and RSM established good correlations with the experimental data but ANN showed higher predictive supremacy than RSM based on its higher values of R and lower error indices. Also, ANN-GA provided more economical optimal DESO extraction route. The physico-chemical characteristics, functional groups and fatty acid compositions of the seed oil compared with literature values and suggest high commercial values for DESO. Therefore, the obtained results present a viable method to harness the useful and highly potential seed oil from for many industrial applications.